The Impact of Stress and Emotional Stimulation on Glucose Prediction: An LSTM-Based Approach

Abstract

In this study, the effects of stress and emotional stimulation levels on blood sugar estimation were investigated. Continuous glucose monitoring (CGM), Galvanic Skin Response (GSR) and skin temperature data of 12 patients in the OhioT1DM dataset were used. Accordingly, three different scenarios were created as glucose data only, glucose and GSR data, and glucose and skin temperature data. In each scenario, the LSTM model was trained and the results were analyzed. During model training, predictions were made for 5, 10, 15 and 30-minute future horizons using past 30-minute data. As the prediction horizon expands, a decrease in model performance is observed; It was determined that the use of glucose and GSR data together did not provide sufficient effect in glucose estimation, and the use of glucose and skin temperature data together increased the performance of the model especially in short-term prediction horizons and reduced error values.